Stanford, California, United States
• Running thermal simulations to accurately predict thermal performance using advanced CFD software for electronics cooling applications. • Supporting design and implementation of thermal solutions to maximize performance while balancing cross-functional interests. • Defining and guiding thermal experiments to characterize thermal materials, components, and systems. • Analyzing thermal test data, preparing reports, and communicating results to cross-functional teams and management.
Most recently: • Did data-driven & physics-informed uncertainty quantification for RANS simulations of turbulent flows. • Predicted local perturbation strength from mean flow features for data-driven eigenvalue perturbations. • Computed weights from random classification forest votes for data-driven baseline simulation. • Connected OpenCV library to flow solvers (OpenFoam, Fluent) for online use of data-driven perturbation framework using C and C++. Previously: • Studied ultrasound induced sonoporation in the context of enhanced drug delivery. Ran 3d viscous two-phase simulations of cavitating microbubbles in an ultrasound pressure eld next to a cellular membrane. Did structural analysis of the resulting membrane deformation and prediction of membrane rupture. • Worked on data-driven momentum forcing for enhanced RANS predictions. • Experimentally analyzed a jet-in-crossflow using Magnetic Resonance Imaging techniques. • Ran DNS of particle-laden flow with focus on two-way coupling.
• Trained convolutional neural networks to learn differentiable geometry representation based on signed distance function in Pytorch • Implemented loss functions capitalizing on solution characteristics, encoder-decoder networks, transfer learning, training data generator. • Presented work at Ansys TechCon 2020 conference.
• Course assistant for graduate level fluid mechanics class (ME 351A). • Helped design homework and exam problems. • Held office hours.
• Developed methodology for exhaust gas valve simulation using STAR-CCM+. • Characterized valves experimentally and compared results with simulations. • Automated workflow for computational valve characterizations: reduced required working time roughly from one week to one day per valve.